１

Strengthening International Governance and Cooperation to Promote the Benign Development of Artificial Intelligence

Oey studysessionof thePoliticalBureau of theCPC Central Committee：“Artificial intelligence （AI）， as a strategic technology leading a new round of technological revolution and industrial transformation， is profoundly changing human production and lifestyles.” AI technology has greatly liberated human mental capacity， fundamentally transforming the paradigm of knowledge production and reshaping the landscape of human civilization at anunprecedented speedand scale. It is profoundly influencing the global economic， political，and military landscape aswell as the development of human society， increasingly drawing high attention from governments， industries，and academia worldwide. Incontrast，theadaptationof theAI governance system has significantly lagged behind the breakthrough progress of the technology， making international AI governance issues increasinglyprominent. Therefore，we need to accurately grasp the global development trends of AI， scientifically assess the risks and challenges of AI development， strengthen international governance and cooperation， and promote AI development ina benign direction for the benefit of humanity.

ACCURATELYGRASPINGTHEGLOBAL DEVELOPMENTTRENDOFARTIFICIAL INTELLIGENCE

Sinceitsinceptionin 1956，AI has undergonenearly7O yearsofupsand downs. In recent years， generative AI technologies represented by ChatGPT， Sora，andDeepSeekhaveachieved continuousbreakthroughs，sparkinga new wave of AI enthusiasm globally. Overall， thedevelopmentof AI features the following basic trends：

I.Specialized AI Is Increasingly Maturing

SpecializedAIreferstoAIintended for specific fields，with singular tasks，cleardemands，andwell-defined applicationboundaries，therefore achieving breakthroughs in respective fields.In single-item tests of partial intelligentlevels，itcansurpasshuman intelligence.For example，biometric featuressuchasfaces，irises，and gaits can berecognized withhigh precision and efficiency，and have been successfully applied in multiple fields includinglawenforcement，civil affairsand disasterrelief;AI-assisteddiagnostic systemshavecovered over 3，000com mondiseasesandareusedin more than 27，oo0 primary healthcare institutions;Tesla'sfullyautonomous drivinghasachieved the firstend-toend AI-driven autonomous driving， among others.

II.LargeModelsHave Achieved Significant Breakthroughs

Largemodelsrefer tomachine learning models with ultra-large scaleparametersandcomplexcomputational structures.As the most attention-grabbingdirection in the current AI field， large models are increasingly becoming strategic infrastructure.Leveraging the develop mentparadigmof“pre-trained large models + fine-tuning for downstream tasks\"，training costs and development cyclescanbeeffectivelyreduced to enable scale production. Since the releaseofChatGPTattheendof2022， largemodelshave entereda periodof explosive growth， acquiring capabilities for multi-modal understanding and multi-type content generation. At thebeginningof2025，DeepSeek-R1 emerged， offering advantages such as open-source innovation and low computational costs，making large models truly accessible to the general public.

III. Generative AI Is Thriving

Generative AI refersto AI technologythat automaticallygeneratestext， images，audio，video，code，andother content based onalgorithms，models， andrules.Compared to traditional AI systems thatcan onlyprocessbased on input data， generative AI systems cancreatenewcontentontheirown. Driven by large models， generative AI hasdevelopedrapidlyandiswidely applied in professional Qamp;A，programming，poetry creation， painting， and other areas，demonstrating powerfulgenerativecapabilitiesthathave rocked the industry. For example， ChatGPTwas included in theNature's10 in 2023，marking the first-ever inclusion of a computer program in the list.

IV.EmbodiedAIandHumanoid Robots Are Gaining Attention

EmbodiedAIreferstotheintelligence seekingcontinuousgrowth through the physical entity's ongoing interactionwith the external environment.Itcanovercomethelimitations of traditionalAI，whichreliesonstacking“computational power + data\" andlacksadaptabilitytoscenarios， providing a new perspective for exploringthemechanismsofhuman intelligence formation and a powerful engine for translating scientificand technological achievementsintorealworld productivity.Humanoid robots， asanimportantcarrierofembodied AI， provide an ideal platform for studying the emergence of intelligence. Today，researchon embodiedAI has enteredafasttrackandwillfurther drivetheparadigm shift ofAI from virtual cognition to physical intelligence through the closed-loop iterationof\"data-algorithm-carrier\".

V.AI-driven Scientific ResearchIs DevelopingRapidly

Over the past few centuries， scientific research has undergone transformationsfromexperimental observation，theoretical analysis， computational simulation，to dataintensiveparadigms.Currently，AIdriven scientific research is leading a newwaveoftransformation，which will fundamentally change the way scientificdiscoveriesaremade. For example，AlphaFold2 can accurately predict the 3D structure of proteins， while AlphaFold3 can predict the structureandinteractionsofalllife molecules with unprecedented precision， bringing revolutionary changes to biomedical research. Additionally， AI hasbeenwidely applied in vaccine development，chipdesign，weather forecasting，and manyother fields.

VI. AI Is Accelerating Its EmpowermentofVarious Industries

Asastrategic technology leading anewroundoftechnologicalrevolution and industrial transformation， AIisprofoundlychanging human productionandlifestyles.AI isaccelerating its deep integration with the real economy，enhancing social productivity，promoting the transformation and upgradingof traditionalindustries， fostering the vigorousdevelopmentof emerging industries， cultivating new highlandsforfutureindustries，and driving the rapid growth of the“unmanned economy\".Meanwhile，AI is having a significant impact on social lifedomains such as transportation， home living，education，healthcare， and elderly care，providing better solutions for promoting social equity， enhancing people'swell-being，and improving the qualityand happiness ofhuman life.

VII.The Social ImpactofAI Is Becoming Increasingly Prominent

Alongwiththerapiddevelopment of AI，its social impact is diverse and all-encompassing.On the one hand， AI brings many positive effects，such as boosting the economy，serving people's livelihoods，and benefiting society;on the other hand，italso brings many social problems，such as securityrisks，legal uncertainties，ethical anomalies，and privacybreaches.For example， themisuse of generative AI technologyto create 1 information fordefamation，extortion，and maliciousattacks poses significant security risks;theabilityofsmartdevicesand algorithmstoexploreandanalyze personalprivacyfarexceedsthat of previous technologies，makingpersonal informationand privacyissues including personal trails，consumption data， biometric features increasingly complexand difficult toprevent.

VIII . General AI Remains a Longterm Goal

General AI aimsto developmachines with the samelevel of intelligence as or even higher than that of humans，and may even enable machines to have autonomous conscious ness.Inrecent years，as the capabilities of neural network models continue toimprove，thepublichasdeveloped overly high expectationsupon AI， and some experts and scholars have madeunrealisticpredictionsabout the developmentof general AI. Currently， AI still hasa significant gap from thelevel of human intelligence，with obvious deficiencies in semantic understanding，reasoning，planning，gen eralization，and continuous learning. Italso lacks an essential understanding of the spatiotemporal operational lawsof theobjectivephysical world and lacks deep empathy.There is still a longway togo beforeachieving true general AI.

SCIENTIFICALLYASSESSINGTHE RISKSANDCHALLENGESOFAI DEVELOPMENT

General Secretary Xi Jinping pointed out：“AI bringsunprecedented development opportunities and also unprecedented risksandchallenges.” While AI accelerates the profound integration of the digital and physical worlds，italso blurs theboundaries between the virtual and thereal to some extent.Its inherent uncertainties， opaqueness，andlackof explainabilityarebecomingincreasinglyprominent， posingnumerousrisksand challenges to human society.

I.Risksand Challenges to Nation al Security

The first is ideological security risks.Erroneousideological trends such as historical nihilism and populism may spread incyberspace through algorithmicrecommendation technologies，impactingideological security. The second isnational governance environmentsrisks.AI can create1 informationand confuse publicperception through deepfake technologies，and use precise algo rithmsto exacerbate the“information cocoon\"effectand enable opinionmanipulation，leading to aweakening of national explanatory power. The third ismilitarysecurityrisks.Onceatechnological “generation gap”is formed inmilitaryintelligence，technological powerhouses are more likely to engage in political blackmail and technological surprise attacks against developing countries.The cross-domain military applications ofAI in nuclearand space domains will also pose significant risks to military security.

II.Risks and Challenges to Indus. trial Development

The maturity ofAI's technical capabilitiesand product forms，the determination of core application scenariosand thedegreeof industrial integration areall closely related to the development of industries. Achieving mature commercial applications is not something that can be accomplished through short-term massive investmentin fundsand resources; the path is still long： Taking ChatGPTas an example，its initial investment cost was aboutUSD 問問21800million，with daily electricity costs of approximately USD 50，oo0 and per-answer costs of a few cents.Faced with such high costs， for one thing，resource monopolization issues are becoming increasingly prominent， with tech giants forming \"digitalhegemony”thatcontinuously squeezes the survival space of small and medium-sized enterprises.For another， many enterprises and startup teams arerushingin，and some governmentsare also eager to follow suit. Blindly rushing in has rendered the developmentofAI disorderedand resource-wasting：Additionally，some enterprises exploit the AI trend for speculative trading，exacerbating market bubble risks.

III.Risks and Challenges to Social Governance

The first is data privacy leakage risks.Thetechnicalcharacteristicsand application scenarios ofAI determine thatitcan easilyaccess largeamounts of private information when being used，potentially collecting personal trails，browsing records，consumption data，biometric features，etc.，without users'knowledge，posing severe challenges to personal information and privacy protection.The second ismaliciousattack risks.Criminals usedeepfake technologies to create 1 videos and images for extortion， maliciousattacks，and otheractivities， threatening social securityand stability，personal reputation，and property. The third is information pollution risks.The misuse of AI technologies hasledtotheproliferationof1 andlow-quality content on the internet.This“pollution”of information distortsmodels'perceptionof reality， causing irreversibledefectsinresults.

IV.Risksand Challenges to Ethics andtheLaw

First，the development of AI technologies may trigger significant changes to social employment structures.AI can autonomously complete highly complex tasks， greatly reducing the barriers and costs of content production and interaction. Allindustries face the risk oflabor displacement by AI， which may cause unemployment panic and social disorder. Industry boundaries will further blur， disrupting the long-standing social division of labor system.Second， algorithmic bias and machine discriminationmay exacerbate.AI algorithms inherently have limitations in transparency and robustness，which will exacerbate biases or discrimination in generated content， amplifying social injustice. Third， technological ethics issues may be triggered. Technological ethics issues such as digital twins， brain-computer interfaces，and human-machine symbiosis involve risks to human life safety，personal privacy， accountability，social fairness，and justice. The widespread application of humanoid robots will also bring many new issues to a social system where humans are the sole subjects，such as the legal status of“AI companions\". Fourth，the legal system face the risk of failure.The legal subject status of AI systems isquestionable，and the attribution of rights and responsibilities is unclear， leading to problems that traditional laws can hardlyregulate. For example， the intellectual property rights of AI-generated content are difficulttodefine，trafficlawsforautonomous driving systems still need improvement，andtheallocationof liability for lossescaused byAI is hard to determine， among others.

MeetingDavos 2025 OnJanury5inswitrld Secretary-GeneralAntonioGuterresstatedatthe WorldEconomic ForumAnnualMeetingDavos 2025thatclimatechangeand theuncontrolled expansion ofAI have been threateningthe survival and developmentof humankind.

ACCELERATINGINTERNATIONALGOVERNANCE ANDCOOPERATIONON AI

Inrecentyears，the subjectsand structures of global AI governance havebecome increasinglydiversified， exhibiting an overall trend of comprehensivedomains，multi-levels，and broad cooperation.Various initiatives， legislation，and ruleshaveemerged in a surge， covering politics， economy， law，healthcare，sociallife，andother fields.Undertheframeworkofthe UnitedNations，countriesand variousinternational organizations have activelyimplemented AI governance and regulation， while industry organizationsand enterprises have also participated in promoting self-regulation. Regional cooperation，multilateral cooperation，and bilateral cooperation have continued to strengthen. In 2023， China issued the Global AI Governance Initiative， proposing constructive solutionstowidely concerned issues of AI development and governance， pro vidingablueprintforrelevantinternational discussions and rule-making， and demonstrating China's sense of responsibilityasamajorcountry.

However，currentinternational AI governance still faces multiple challenges.Thefirstisthelackofaunified global governance framework. There is stillalack of consensus on major issues among countries，and manyregulatorydifferencesaredifficult to coordinate. Non-binding internationalinitiativeshavelimited effectiveness，andtheconstruction ofgovernance ruleswith substantive bindingforcestill needstime.The second is divergent interests among countries.Developed economiesare more concerned with technologicalsecurityand ethical issues，while emergingandothereconomiesemphasize economic and social develop mentopportunities.The imbalanceof governancepowerfurtherexacerbates the divergence of focus. The third is the fragmented governance. The rapid development of AI technology and the lagin theconstruction of the governancesystemcreateacontradiction.Static regulationscan hardly addressdynamicrisks，whichmeans afragmentedgovernanceframework with low efficiency.

President Xi Jinping emphasized the need to prioritize people and ensureAI serves good causes，strengthening AI rules governance within theframeworkof theUnitedNations. Specifically， accelerating international governanceandcooperationonAI requires breakthroughs in the following fourareas：

I. Upholding“People-Oriented\" Principle and Adheringto the Correct Direction for AI Development

No matter how technology evolves， we must stay true to our original aspirations and pursue benign development in the intelligent era. On the one hand， multilateral governance entities should further reach a value consensus on AI development，establish fundamental governance principles，and aim to enhance human common well-being and serve people's comprehensive development on the premise of ensuring social security and respecting human rights and interests， ensuring that AI always develops ina direction beneficial to human civilization. On the other hand， the international community should form a risk consensus on AI development and provide guidance at the legal and ethical levels， establishing an AI risk assessment and early warning systemas soon aspossible toensure that AIresearch and development and its application activities are safe， controllable，andalignedwithsharedhuman values，preventing harm tonational security，public interests，and the legitimaterightsand interestsof organizationsand individuals.

II.Adhering to“Openness and Sharing”PrincipleandPromotingInclusiveand Accessible AI Applications

Due to differences in scientific and educational capabilities，infrastructure，industrial structures，and socio-economic foundations among countries and regions，the issue of imbalanced AI developmenthas come to the fore. On the one hand，we should deepen open-source initiatives，promote the sharing of AI technologies， talent，and infrastructure，help Global South countries strengthen their technological capabilities，establish an international AI technology sharing platform，break data barriers and technological monopolies，and enable fair access to dataresourcesand shared AI benefits，so as to bridge the global AI divide.On the otherhand，we should continuously expand exchanges and cooperation among relevant academic organizations，enterprises，andindustry associations，building an open and win-win international scientific and technological cooperation ecosystem， striving to break restrictions imposed by certain countries on scholars from othercountries to participate in international exchanges and cooperation， andavoiding technological blockades and talent protectionism that hinder technological exchangesand progress.

III.Adhering to“Governance Through Technology”Principle and InnovatingAI Governance Methods

The decision-making processes and logic of AI are opaque，exhibiting “blackbox”characteristics.Current technologieslackmechanismsforselfregulationand control，creatingaclear governancevacuum atthe technical level.Researchanddevelopment entitiesin various countries and regions should continuously improve the explainabilityandpredictabilityofAI， creating auditable，supervisable， traceable，and reliableAI technologies. They should treat computational power， the mostintuitiveand quantifiablemetric， as the object of governance，indirectly achieving AI governance.At the same time，they should use AI technologies to prevent AI risks，actively designing and developing technologies and applications that can effectively govern AI， trackingand evaluating the imple mentation ofAI governance frameworks，and enhancing the technical capabilities ofAI governance.

IV.Adhering to “System Integration”and Buildinga Multilateral AI Governance Framework

Againstthe backdropofdiverse challenges to world peace and development， the international community should build broad consensus globally，actively construct an open，fair， and effective multilateral governance mechanism，and jointlypromote AI technologiesto trulybenefithumanity.On the one hand，countriesand regions，aswell asrelevantinternational organizations，should strengthen the alignment and coordination ofgovernancerulesand technical standards， reach consensus onkey issues such as AIsafety，ethics，data privacy，crossborder regulation，and military applications，establish and improve laws andregulations，andrefineAIethical guidelinesand accountabilitymechanisms.They should fullyrespect nationaldifferences，enhancethevoice of developing countries，and ensure equalrightsand opportunitiesforall countriestodevelopand utilize AI technologies.On the otherhand，other diverse governance entities such as enterprises，researchinstitutions，social organizations，andindividualsshould activelyplayrolesmatching their own identities，jointlyparticipatingin the constructionand improvementof the AI governance system，achieving the development ofAI across the globe featuring extensive consultation， joint contribution，and shared benefits.

International governance of AI is anissue of the times concerning the promotion of building a community withashared future formankind.Facing the opportunities and challenges brought byAI，we need both mutual standard recognition and joint risk prevention，as well as ethical guidance andlegal safeguards.Wemustfurther build consensusand deepen collaboration，alwaysadhering to the“people oriented”value orientation，ensuring that technological progress keeps pace withcivilization development，and makingAI apositive force forenhancing human well-being